Apppublic

OutstandingOm/knowledge-graph-env

sourceHugging Faceupdated 5mo agoView on Hugging Face

1likes

App README

Self‑Evolving Knowledge Graph Environment for Continuous Reasoning

Think of it as a system that stores knowledge like a graph, learns continuously after deployment, and answers in milliseconds.

Fast, memory‑efficient, and continuously learning – designed for real‑world customer support automation.

🎯 What is this?

A reinforcement learning environment that simulates customer support ticket triage. An agent learns to:

1.Identify the main issue from a support ticket.
2.Relate it to a known concept in a dynamic knowledge graph.
3.Answer with a resolution action.

Unlike static environments or large language models (LLMs), our system maintains persistent memory of concepts and relationships and improves after deployment through continuous background updates – without retraining.

🧠 Key Innovations (Why it’s different)

Feature	What it does	Why it matters
Persistent memory	Stores concepts and relationships permanently	The agent never forgets past tickets
Continuous learning	Background trainer updates vectors every 10 seconds	The system gets smarter over time, even after deployment
DNA‑inspired encoding	Each concept is built from structured combinations of simple components (letters A–Z), enabling compact and composable representations. Compact 16‑dim vectors.	Very low memory (~150 MB for 1000 concepts) and fast (<1 ms query)
Reasoning engine	Multi‑hop graph propagation + analogical reasoning	Can answer complex queries like “A is to B as C is to ?”
Deterministic grading	Clear reward rules (exact match → 1.0, partial → 0.7/0.3)	Judges can reproduce scores 100% of the time

⚙️ How It Works (simple version)

1.Input – a support ticket (e.g., “I can’t log in”)
2.Feature extraction – keywords become features (e.g., “login”, “password”)
3.DNA encoding – each feature maps to a sequence of letters (A–Z) with learnable vectors; a concept vector is the sum of its features’ encodings. DNA‑inspired encoding means each concept is built from structured combinations of simple components (letters A–Z), enabling compact and composable representations.
4.Knowledge graph – concepts are nodes; relationships are edges. When two concepts are linked, their vectors move closer – the whole graph learns.
5.Reasoning – FAISS search (similarity) + multi‑hop activation + analogical arithmetic.
6.Reward – deterministic scoring based on exact match, substring, or word overlap (0.0 / 0.3 / 0.7 / 1.0).

📊 OpenEnv Tasks (3 independent graders)

Task	Difficulty	Description	Example input	Expected output
`task_easy`	Easy	Identify the main concept	“Login not working”	`login issue`
`task_medium`	Medium	Find the correct relation	“Bill is wrong”	`refund`
`task_hard`	Hard	Provide the resolution	“Locked out after failed payment”	`reset password`

All graders are deterministic and return a score between 0.0 and 1.0.

⚡ Performance (on 2 vCPU / 8GB)

Metric	Value
Latency per step	< 1 ms
Full episode (3 steps)	< 5 ms
Memory for 1,000 concepts	~150 MB
Determinism	100% (same input → same score)
Scalability	Up to 100,000 concepts with < 2 ms search

🔄 Comparison with LLMs (balanced view)

Aspect	LLM‑based approach	Our environment
Memory	Context window only; external DB needed	Persistent graph, built‑in
Latency	Seconds	Microseconds
Cost	API or GPU	Zero (CPU only)
Learning after deployment	Expensive fine‑tuning	Automatic background updates
Best for	General reasoning, creative generation	Structured, repetitive, fast queries

We do not claim to replace LLMs – we provide a complementary solution for tasks that require low latency, persistent memory, and incremental learning.

🚀 Real‑World Use Cases

—Customer support ticket routing – learn new issues continuously.
—Enterprise knowledge management – keep a living graph of documents.
—Educational tutoring systems – track student misconceptions.
—Legal case law analysis – link new precedents to old rulings.

🛠️ How to Run

Locally

bash

git clone https://github.com/outstandingom/dna-modal.git
cd dna-modal
pip install -r requirements.txt
python inference.py